Tom Griffiths Invited Talk
in
Workshop: UniReps: Unifying Representations in Neural Models
Why should we care if humans and machines have similar representations?
Abstract:
A substantial amount of work has shown that there is a strong correspondence in the representations acquired by humans and machines, particularly in the case of vision and language. I will summarize some work from our lab supporting this idea, and then turn to the next question: why should we care? I will present recent work showing that closer correspondences between representations in humans and machines can support more rapid learning by those machines, and that misalignment can be hazardous.
Chat is not available.